Automatic Music Genres Classification using Machine Learning

Muhammad Asim,Zain Ahmed

doi:10.14569/ijacsa.2017.080844

Abstract

Classification of music genre has been an inspiring job in the area of music information retrieval (MIR). Classification of genre can be valuable to explain some actual interesting problems such as creating song references, finding related songs, finding societies who will like that specific song. The purpose of our research is to find best machine learning algorithm that predict the genre of songs using k-nearest neighbor (k-NN) and Support Vector Machine (SVM). This paper also presents comparative analysis between k-nearest neighbor (k-NN) and Support Vector Machine (SVM) with dimensionality return and then without dimensionality reduction via principal component analysis (PCA). The Mel Frequency Cepstral Coefficients (MFCC) is used to extract information for the data set. In addition, the MFCC features are used for individual tracks. From results we found that without the dimensionality reduction both k-nearest neighbor and Support Vector Machine (SVM) gave more accurate results compare to the results with dimensionality reduction. Overall the Support Vector Machine (SVM) is much more effective classifier for classification of music genre. It gave an overall accuracy of 77%.

Highlights

Nowadays, a personal music collection may contain hundreds of songs, while the professional collection usually contains tens of thousands of music files
Most of the music files are indexed by the song title or the artist name [1], which may cause difficulty in searching for a song associated with a particular genre
The recognition rate is improved from 55% to 64% which is equivalent to the previous recognition rate of Support Vector Machine (SVM), it shows that with all features k-nearest neighbor (k-NN) is more effective classifier

Summary

Introduction

A personal music collection may contain hundreds of songs, while the professional collection usually contains tens of thousands of music files. Most of the music files are indexed by the song title or the artist name [1], which may cause difficulty in searching for a song associated with a particular genre. With large music database the warehouses require an exhausting and time consuming work, when categorizing audio genre manually. Music has been divided into Genres and sub genres on the basis on music and on the lyrics as well [2]. To make things more complicate the definition of music genre may have very well changed over time [3]. Proceedings - IEEE International Conference on Multimedia and Expo, 881–884

Objectives

Methods

Findings

Conclusion